De Aitalkx Dms Rag 2026 05 17
Nicholas Michael Kloster / NuClide Research nicholas@nuclide-research.com
2026-05-17
Re: Unauthenticated Elasticsearch with AI/RAG workload, AItalkx (aitalkx.com). DMS RAG / document workflow
IP / Host: 84.247.189.64 (cluster docker-cluster (372e2468a9d2))
Severity: HIGH
I’m an independent security researcher conducting good-faith AI infrastructure research under the NuClide Research umbrella (CISA disclosures CVE-2025-4364, ICSA-25-140-11). This is an unsolicited coordinated disclosure. No engagement exists with your organization. I have not read, modified, or exfiltrated documents. Only index metadata (schema, counts, sizes) needed to identify the exposure.
Summary
The Elasticsearch endpoint at http://84.247.189.64:9200 is reachable from the public internet with no authentication. The cluster carries one or more AI / RAG workloads (dense_vector or knn_vector fields). TLS SAN includes aitalkx.com.
State (verified 2026-05-17): MID-WIPE: extortion marker planted, DMS document corpus still alive
Infrastructure
| Field | Value |
|---|---|
| IP | 84.247.189.64 |
| ES version | OpenSearch 2.19.1 |
| Cluster name | docker-cluster (372e2468a9d2) |
| Country | Germany |
| Hosting | Contabo GmbH (registrar: Vautron Rechenzentrum AG) |
Alive on the cluster
| Index | Docs | Size | Notes |
|---|---|---|---|
dms_data | 6,762 | 6.3 MB | primary document store |
dms_vectors | 1,727 | 30.4 MB | knn_vector 768d — DMS document embeddings |
n_general_document_generaldocument | 258 | 785.6 KB | general documents |
n_general_document_dms_purchase_order_workflow | 58 | 87.8 KB | purchase-order workflow documents |
n_general_folder_generalfolder | 106 | 86.9 KB | folder hierarchy |
n_general_folder_workspace | 59 | 53.3 KB | workspaces |
dms_documentvectors | 0 | 208 B | knn_vector schema, empty |
Extortion marker
An index named read_me is present on the cluster. This is the calling card of the Meow / Indexrm extortion campaign, an automated wipe-and-ransom operation. The marker indicates the attacker has already enumerated your cluster and either has wiped the data or is about to.
Sample marker content typically contains a ransom note pointing to a Bitcoin wallet and an email channel (wendy.etabw@gmx.com or one of two clone actor channels, scandal@onionmail.org, db-recovery@sharebot.net). We have separately reported all three channels to abuse contacts.
Do not pay. Our wallet-blockchain analysis (mempool.space) shows only ~5 victims across thousands of marked hosts have paid; the wallet has received roughly 0.018 BTC against a population of 4,400+ marked instances. Paying does not get your data back. The campaign is wipe-first, ransom-as-afterthought.
Why it matters
Your DMS RAG corpus is fully exposed: 6,762 document records, 1,727 knn_vector embeddings (dms_vectors, 768d), plus purchase-order workflow documents. The ‘read_me’ marker is in place. Take it offline before the attacker completes the wipe.
One-line fix
Bind the Elasticsearch HTTP listener to the loopback interface and require authentication for any non-local client. For a Docker deployment:
docker run -p 127.0.0.1:9200:9200 -e xpack.security.enabled=true elasticsearch:2.19.1If the cluster is on a host network: set network.host: 127.0.0.1 in elasticsearch.yml, restart, then put it behind a reverse proxy or VPN with authentication.
For OpenSearch clusters: enable the security plugin with plugins.security.disabled: false.
Reference
Companion case study with the full 22-host survey: AI-LLM-Infrastructure-OSINT/blob/main/case-studies/commercial/22-ai-stack-attribution-2026-05-17.md
Companion methodology notes:
- Snapshot vs delta measurement (Insight #29)
- WHOIS over slug heuristics for disclosure routing (Insight #04)
Happy to provide additional forensic detail if useful. No response is required.
Regards, Nicholas Michael Kloster / NuClide Research nicholas@nuclide-research.com